Identification method and device based on biological image identification model, equipment and medium

Abstract

The invention discloses an identification method and device based on a biological image identification model, equipment and a medium, and generally relates to the technical field of computers. The method comprises the following steps: performing biological object calibration segmentation processing on a target biological image through a segmentation model to obtain a target biological feature which is used for representing a to-be-identified biological object in the target biological image; performing abnormal factor detection on the target biological characteristics through a detection model to obtain an abnormal detection result of the to-be-identified biological object; the anomaly detection result comprises whether the to-be-identified biological object has an abnormal factor or not; the abnormal factors comprise at least one of the abnormal size of the biological object, the abnormal form of the biological object and the abnormal absorption characteristic of the biological object to ray energy; and performing property judgment on the abnormal factors of the target biological characteristics through a characteristic judgment model to obtain the properties of the abnormal factors. The training cost is greatly saved while the training effect is ensured.

Description

Technical Field

[0001] The present disclosure generally relates to the technical field of image classification models, and more particularly to a recognition method, device, equipment and medium based on a biological image recognition model. Background Art

[0002] As technology continues to advance, deep learning technology for image recognition is also developing.

[0003] In the related art, the biological image recognition process is a method of performing image recognition by using a learning model. Specifically, the process is to train the model with a large number of biological images with target feature information to obtain a model that can recognize biological images, and then use the model for training.

[0004] However, in practical applications, since the images that ultimately need to be recognized by the model are flexible and changeable, if the training samples are not sufficient or rich enough, the model training effect may be poor and the image recognition may often be wrong during the actual image recognition process. If a model is to be able to recognize accurately changing images, the training cost will be too high and the final model will be too complex and have a poor fit. Summary of the invention

[0005] In view of the above-mentioned defects or deficiencies in the related art, it is desired to provide a recognition method, device, equipment and medium based on a biological image recognition model, which can solve the problem of achieving a balance between training cost and training effect of the current biological image recognition model, while ensuring the training effect, greatly saving training costs.

[0006] In a first aspect, a recognition method based on a biological image recognition model is provided, the method comprising: Performing biological object calibration and segmentation processing on the target biological image through the segmentation model to obtain target biological features, wherein the target biological features are used to characterize the biological object to be identified in the target biological image; The target biological feature is subjected to abnormal factor detection through the detection model to obtain an abnormal detection result of the biological object to be identified; the abnormal detection result includes whether the biological object to be identified has abnormal factors; the abnormal factors include at least one of abnormal size of the biological object, abnormal morphology of the biological object, and abnormal absorption characteristics of the biological object to radiation energy; The abnormal factors of the target biological features are judged by the feature judgment model to obtain the properties of the abnormal factors, which are used to characterize whether there is a risk in the abnormal factors and the risk level of the abnormal factors.

[0007] In this application, the target biological image is first calibrated and segmented by a segmentation model to obtain the target biological feature (the above target biological feature is used to characterize the biological object to be identified in the target biological image); then, the target biological feature is detected by a detection model for abnormal factors to obtain the abnormal detection result of the biological object to be identified, and the abnormal detection result includes two results: the presence of abnormal factors in the biological object to be identified or the absence of abnormal factors. The abnormal factors include at least one of the abnormal size of the biological object, the abnormal shape of the biological object, and the abnormal absorption characteristics of the biological object to the radiation energy; finally, the abnormal factors of the target biological feature are judged by the feature judgment model, and the nature of the abnormal factors is obtained, which is used to characterize whether the abnormal factors have risks and the risk level of the abnormal factors. In this way, the target biological graphics are gradually identified and decomposed by multiple different models, and the abnormal factors in the target biological image and the risk level of the abnormal factors are finally determined, so that the amount of data and data complexity required for each model to identify are greatly reduced, and the pertinence of the identification is enhanced. While ensuring the quality of model processing, the speed of each model in identifying the target biological image is greatly improved, that is, the accuracy and efficiency of the model in identifying the target biological image are greatly improved.

[0008] In a second aspect, a recognition device based on a biological image recognition model is provided, comprising: An execution module, configured to perform biological object calibration and segmentation processing on a target biological image through a segmentation model to obtain a target biological feature, wherein the target biological feature is used to characterize a biological object to be identified in the target biological image; The execution module is further used to perform abnormal factor detection on the target biological feature through the detection model to obtain an abnormal detection result of the biological object to be identified; the abnormal detection result includes whether the biological object to be identified has an abnormal factor; the abnormal factor includes at least one of an abnormal size of the biological object, an abnormal morphology of the biological object, and an abnormal absorption characteristic of the biological object to radiation energy; The execution module is also used to determine the properties of the abnormal factors of the target biological features through a feature determination model to obtain the properties of the abnormal factors, and the properties are used to characterize whether the abnormal factors have risks and the risk level of the abnormal factors.

[0009] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in the first aspect is implemented.

[0010] In a fourth aspect, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that when the program is executed by a processor, the method described in the first aspect is implemented.

[0011] In a fifth aspect, a computer program product is provided. The computer program product includes instructions, and the instructions are executed by a processor to implement the method described in the first aspect.

[0012] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings: Figure 1 A schematic diagram of a process flow of a recognition method based on a biological image recognition model provided in an embodiment of the present application; Figure 2 A schematic diagram of the execution process of the recognition method based on the biological image recognition model provided in the embodiment of the present application; Figure 3 A schematic diagram of image preprocessing of a recognition method based on a biological image recognition model provided in an embodiment of the present application; Figure 4 A schematic diagram of the structure of a recognition device based on a biological image recognition model provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0014] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0015] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0016] In recent years, deep learning technology has made significant progress in various fields, including biometrics. For example, in identifying biological tissue morphology, it can extract key image content that users are concerned about from the acquired biological images and analyze the nature of the key image content, thereby greatly improving the speed and accuracy of identifying key image content of biological tissue morphology.

[0017] There are several ways to use learning models for image recognition: 1) By preprocessing the image and multi-resolution input data, a 3D dual-channel compressed excitation deep learning model based on a dual-channel network and a compressed excitation network is constructed. However, the data set collected by this method is actually very limited, and the model complexity is high. Therefore, there may be overfitting problems when dealing with complex scenes; 2) A model for recognizing image data is obtained by training a random forest model. Although this model can handle a large number of feature variables after training, it is difficult to select suitable feature variables for training in the actual training process. If the selected feature variables are not reasonable, the performance of the model obtained by the final training is also poor, and the technical effect of accurately recognizing images cannot be achieved; 3) Image recognition is performed by combining a constructed entity relationship joint extraction model with a regression analysis function. Specifically, this method extracts feature information from basic image information to train an image recognition prediction model. However, this method is essentially weak in generalization, and the model finally trained cannot recognize feature information that has slightly changed in other images.

[0018] From this, it can be seen that in the above-mentioned practical application process, since the images that ultimately need to be recognized by the model are flexible and changeable, the trained model may fail in the actual image recognition process if the training samples are not sufficient or rich enough. If a model that can accurately recognize changing images is desired, the training cost will be too high, resulting in a final model that is too complex and poorly fitted.

[0019] Specifically, the above related technologies have the following problems: 1) Many existing models are based on time series methods, which tend to ignore contextual information, resulting in reduced judgment accuracy. For example, the number of key features in the image will be inaccurate, ultimately leading to a high misjudgment rate.

[0020] 2) Insufficient image preprocessing, the noise in the original image itself will affect the accuracy of discrimination.

[0021] 3) Existing image recognition models have high requirements for computing power, which will increase the user's usage costs and limit the promotion of image recognition using models.

[0022] 4) Existing data resources are not expanded enough and data resources are not fully utilized.

[0023] 5) The distribution of the given positions of various biological features that may exist in biological images is unclear, so it is impossible to implement a multi-resolution pixel positioning method in the image recognition process.

[0024] 6) Low risk of overfitting and automatic recognition of image features.

[0025] Based on this, the present application proposes a recognition method, device, equipment and medium based on a biological image recognition model, which can solve the problem of achieving a balance between training cost and training effect of the current biological image recognition model, while ensuring the training effect, greatly saving training costs.

[0026] Figure 1 FIG. 1 is a flow chart of a recognition method based on a biological image recognition model provided by an embodiment of the present application. Figure 1 As shown, the method includes the following steps 301-303: Step 301: Perform biological object calibration and segmentation processing on the target biological image through the segmentation model to obtain the target biological features.

[0027] Exemplarily, the above segmentation model can be a neural network segmentation model.

[0028] In one embodiment, the neural network segmentation model may be a U-net model.

[0029] Exemplarily, the above segmentation model can be used to identify and determine the boundaries of biological objects in the target biological image, and then perform segmentation processing to obtain target biological features in the target biological image.

[0030] It can be understood that the above segmentation model is a pre-trained model specifically used to identify target biological features from target biological images. The training process can refer to the subsequent content and will not be described in detail here.

[0031] Furthermore, for the segmentation model, the biological image it receives can be one biological image or multiple biological images. When there is one biological image, the above-mentioned target biological image is the one biological image; when there are multiple biological images, the above-mentioned target biological image can be one of the multiple biological images, or it can be multiple or all of the multiple biological images. The embodiments of the present application do not limit this.

[0032] In the embodiment of the present application, the above-mentioned target biological image can be a photo or other biological recognition image, and the embodiment of the present application does not limit this.

[0033] In the embodiment of the present application, the above-mentioned target biological feature is used to characterize the biological object to be identified in the above-mentioned target biological image.

[0034] It can be understood that the above-mentioned target biological image may include multiple biological features, and the target biological feature may be a part of them.

[0035] In an example, the target biological feature may be an independent and complete biological feature in the target biological image and / or a biological feature that occupies the largest area ratio in the target biological image.

[0036] Exemplarily, the segmentation process may be to calibrate the boundaries of all biological objects in the target biological feature, determine the boundary position of each biological object, and then perform segmentation on it.

[0037] Example 1: Figure 2 As shown, assuming that the above target biological image is an image representing human chest tissue, the image boundaries of biological objects that may be included in the human chest include: the boundary of the lung image, the boundary of the heart image, the boundary of part of the bronchial image, and the boundary of part of the blood vessel image. Then the above target biological feature is the biological feature that is independent and complete in the image representing human chest tissue and occupies the largest area ratio in the target biological image, that is, the lung image feature. The image boundaries of each of the above biological objects are calibrated through the neural network segmentation model (that is, the above segmentation model), and then image segmentation processing is performed, and finally the above lung image feature, that is, the target biological feature, is segmented.

[0038] Step 302: Detect abnormal factors of the target biological features through the detection model to obtain abnormal detection results of the biological object to be identified.

[0039] In the embodiment of the present application, the above-mentioned abnormality detection result includes whether there are abnormal factors in the above-mentioned biological object to be identified.

[0040] In the embodiment of the present application, the above-mentioned abnormal factors include at least one of abnormal size of the biological object, abnormal morphology of the biological object, and abnormal absorption characteristics of the biological object to radiation energy.

[0041] In the embodiment of the present application, the above-mentioned abnormal factors are the criteria for determining abnormal detection results.

[0042] It can be understood that the target biometric feature is a complete biological object in the target biological image, while the anomaly detection result represents the presence of anomalies in the target biometric feature, that is, an unconventional biological object.

[0043] Furthermore, the above abnormal factors can be used to characterize factors or features corresponding to the content that should not originally belong to the target biometric feature but currently appears in the target biometric feature. For example, when the target biometric feature is a lung image feature, the above abnormal factors can be factors in the lung image feature that are consistent with the morphology, size, and radiation absorption capacity of lung nodules.

[0044] In the embodiment of the present application, the abnormal size of the biological object, the abnormal shape of the biological object, and the abnormal absorption characteristics of the biological object to radiation energy can be preset or user-defined, and the embodiment of the present application does not limit this.

[0045] In one example, when the abnormal factor is an abnormal size of a biological object, that is, there is a biological object with an abnormal size in the target biological feature, it is determined that there is an abnormal detection result of the biological object to be identified.

[0046] In one example, when the abnormal factor is an abnormal form of a biological object, that is, there is a biological object with an abnormal form in the target biological feature, it is determined that there is an abnormal detection result of the biological object to be identified.

[0047] In one example, when the above abnormal factor is the biological object's response to radiation energy, that is, there is a biological object in the target biological feature that meets the requirements for radiation energy, it is determined that there is an abnormal detection result for the biological object to be identified.

[0048] Exemplarily, the above detection model can be used to detect multiple abnormal factors of target biological features. When there is an abnormal factor in a target biological feature, the abnormal detection result of the biological object to be identified can be determined.

[0049] It can be understood that the above detection model is a pre-trained model specifically used to detect abnormal factors from target biological features. The training process of the detection model can refer to the subsequent content and will not be repeated here.

[0050] Furthermore, for detection, the content of detection is the aforementioned target biological feature, that is, a partial image area in the target biological image.

[0051] Exemplarily, the detection model may be a model constructed by combining a feature pyramid network (FPN) with ResNet50 as the skeleton and a Relu activation function.

[0052] Example 2: Combined with the above example 1, Figure 2 As shown, assuming that the target biological feature is the above-mentioned lung image feature, the abnormal factors include any one of the size feature of the lung nodule, the morphological feature of the lung nodule, and the radiation energy absorption characteristic of the lung nodule. After the detection model detects and identifies the lung image feature, if the lung image feature includes any one of the three features of the size feature of the composite lung nodule, the morphological feature of the lung nodule, and the radiation energy absorption characteristic of the lung nodule, it is considered that the area belongs to the abnormal detection result.

[0053] Step 303: Determine the nature of the abnormal factors of the target biological features through the feature determination model to obtain the nature of the abnormal factors.

[0054] In the embodiment of the present application, the above properties are used to characterize whether the above abnormal factors have risks and the risk level of the above abnormal factors.

[0055] It will be appreciated that the nature of the above-mentioned abnormal factors can be used to determine the degree to which the abnormal factors require attention.

[0056] Furthermore, for some abnormal factors, they are abnormal factors that change normally over time and do not require users to pay too much attention; while for some abnormal factors, the degree of their change is abnormal factors that require users to pay close attention. The nature of the above abnormal factors is used to distinguish between normal abnormal factors that do not require users to pay too much attention and abnormal factors that require users to pay close attention.

[0057] Exemplarily, the feature determination model is a model for users to distinguish different properties of the abnormal factors.

[0058] In one example, the feature determination model may be an EfficientNet network model.

[0059] The following describes the EfficientNet network model: The EfficientNet network model achieves this goal by balancing the depth, width, and resolution of the received image information in order to optimize performance within a predetermined computational budget, that is, it can maximize the rate by balancing the depth, width, and resolution while ensuring accuracy. The EfficientNet network model provides a compound scaling method (scaling all depth, width, and resolution dimensions) that can help the model achieve maximum accuracy improvement. The baseline network has a great impact on the scaling effect of the model. The efficient network can handle a wide range of image classification tasks, and it has become a good model for transfer learning.

[0060] Example 3: Combined with the above example 2, Figure 2 As shown, after the lung nodule area in the lung image features is identified, the nature of the lung nodule area will be determined through the feature determination network, that is, whether it belongs to an area that needs attention or an area that does not need attention.

[0061] In the method provided in the embodiment of the present application, the target biological image is first subjected to biological object calibration and segmentation processing through a segmentation model to obtain a target biological feature (the above-mentioned target biological feature is used to characterize the biological object to be identified in the target biological image); then, the target biological feature is subjected to abnormal factor detection through a detection model to obtain an abnormal detection result of the biological object to be identified, and the abnormal detection result includes two results: the presence of abnormal factors in the biological object to be identified or the absence of abnormal factors, and the abnormal factors include at least one of the abnormal size of the biological object, the abnormal morphology of the biological object, and the abnormal absorption characteristics of the biological object to the radiation energy; finally, the abnormal factors of the target biological feature are subjected to property determination through a feature determination model, and the properties of the abnormal factors are obtained, and the properties are used to characterize whether the abnormal factors have risks and the risk level of the abnormal factors. In this way, by gradually identifying and decomposing the target biological graphics through multiple different models, and finally determining the abnormal factors in the target biological image and the risk level of the abnormal factors, the amount of data and data complexity required for each model to be identified are greatly reduced, and the pertinence of the identification is enhanced, while ensuring the quality of model processing, the speed of each model in identifying the target biological image is greatly improved, that is, the accuracy and efficiency of the model in identifying the target biological image are greatly improved.

[0062] In another embodiment of the present application, a specific implementation method for preprocessing the original image in advance to generate a target biological image is also provided. Exemplarily, the specific implementation of "performing biological object calibration and segmentation processing on the target biological image through a segmentation model to obtain target biological features" mentioned above includes: obtaining original image data, performing image preprocessing on the original image data, and generating a target biological image.

[0063] Exemplarily, the above-mentioned image preprocessing includes at least one of the following: original image denoising preprocessing, original image data preprocessing, original image homogenization preprocessing, original image target data expansion preprocessing, and original image boundary alignment preprocessing.

[0064] Exemplarily, the original image may be an image that has not been processed. It is understandable that, for any model, the target biological image is composed of image data. If the target biological image includes data that is irrelevant to the biological features or abnormal factors or the nature of abnormal factors that need to be identified, it will cause errors in the model recognition process. Based on this, the original image can be pre-processed in advance through image pre-processing to filter out data content that is irrelevant to the image that the model needs to recognize as much as possible, thereby enhancing the accuracy of the recognition process.

[0065] Furthermore, the data preprocessing step is crucial for data normalization because the acquisition process and path of the original images may be different, so the original images may not use exactly the same acquisition protocol. If the original images are directly processed by the model, various errors will occur.

[0066] In an example, the above-mentioned original image denoising preprocessing method may include Gaussian Blur processing, and Gaussian Blur can smooth the image and reduce noise.

[0067] It can be understood that Gaussian blur is a linear low-pass filter that uses a Gaussian function to calculate pixel values. The weight reduction speed is determined by the Gaussian function, so it is called Gaussian blur.

[0068] Furthermore, the algorithm of the above Gaussian blur is as follows: (1) explain x->X coordinate value, y->Y coordinate value, -> mathematical constant PI (value = 3.13), σ->Standard Deviation exp->Exponential function Input: Image I has size m*n, Gaussian kernel Output: Image O has size m*n Normalize G,

[0069] PadtheinputimageIwithzerostoavoidboundaryeffects.Amountofpadding=(k-1) / 2ontheeachside. #Pad the input image I with zeros to avoid edge effects. The amount of padding on each side = (k-1) / 2 Foreachpixel(i,j)intheinputimageI:#For each pixel (x,y) of image I Initialize the output pixel O (i, j) to zero. # Initialize the output pixel to zero.

[0070] Foreachkernelelement(x,y):#Operate for each kernel element (x, y) Computethepixelcoordinates(a,b):#Calculate pixel coordinates

[0071]

[0072] AddtheproductoftheinputpixelI(a,b)andthekernelelementG(x,y)totheoutputpixelO(i,j): #Add the product of input pixel I(a,b) and kernel element G(x,y) to output pixel O(i,j)

[0073] END END Clip the output image to the range [0, 255] #Clip the output image to the range [0, 255] Return the output image O#Return the output image O In an example, the above-mentioned method of preprocessing the original image data may include performing threshold processing on the original image.

[0074] Furthermore, the threshold processing process may include: if the original image intensity is less than a fixed value called a threshold, each pixel in the original image is replaced with a black pixel; if the pixel intensity is greater than the threshold, each pixel in the original image is replaced with a white pixel.

[0075] Furthermore, the algorithm of the above threshold processing process can be: Input: image I of size mxn, threshold Output: Image O of size mxn Foreachpixel(i,j)intheinputimageI:#Each pixel (i,j) in input image I If the pixel value I (i, j) is less than or equal to the threshold value (255): #If the pixel value I (i, j) is less than or equal to the threshold value # (255) settheoutputpixelO(i,j)to255(white)#Set the output pixel O(i,j) to 255 (white) End Else: The output pixel O(i,j) remains the same End End Return the thresholded output image O # Returns the output image O after threshold processing Return the output image O#Return the output image O In an example, the above-mentioned original image homogenization preprocessing method may include: homogenizing the same details and isolated elements in the original image, that is, it is a processing method for extracting small details or isolated pixels in the image.

[0076] Furthermore, the original image homogenization preprocessing method may be corrosion.

[0077] Furthermore, the above corrosion algorithm can be described as follows: Input: size is mxn, structure element B size is kxk Output: Image O of size mxn PadtheinputimageIwithzerostoavoidboundaryeffects.Amountofpadding=(k-1) / 2ontheeachside. #Pad the input image I with zeros to avoid edge effects. The amount of padding on each side = (k-1) / 2 Foreachpixel(i,j)intheinputimageI:#Each pixel in the input image (i,j) Initializeaflagvariableftotrue# Initialize variable flag f to true Foreachstructuringelementelement(x,y):#Enter each structural element element (x,y) Computethepixelcoordinates(a,b):#Calculate pixel coordinates (a,b) (2) (3) If(I(a,b)==0&B(x,y)==1): setftoFalse#Set f to false End If(f==True): settheoutputpixelO(i,j)to1#Set the output pixel O(i,j) to 1 End Else: settheoutputpixelO(i,j)to0#Set the output pixel O(i,j) to 0 End End End Remove the padding from the output imageO#Remove the padding from the output image.

[0078] Return the output image O#Return the output image In one example, the above-mentioned original image target data expansion preprocessing method is a morphological image processing technology, which is used to amplify or expand the details in the original image. Specifically, it can create modified images in the data set corresponding to the original image to expand the size of the training set. The labels of all images are the same, which are used to generate the original image labels of these images.

[0079] Furthermore, the above-mentioned image augmentation algorithm may include the following contents: Input: image I of size mxn, structure element B of size kxk Output: image O of size mxn #Pad the input image I with zeros to avoid border effects. Amount of padding on each side = (k-1) / 2 PadtheinputimageIwithzerostoavoidboundaryeffects.Amountofpadding=(k-1) / 2oneachside Foreachpixel(i,j)intheinputimageI:#For each pixel (i,j) in the input image I, initialize the flag variable f to false Initializeaflagvariableftofalse Foreachstructuringelement(x,y):#For each structural element (x,y) Computethepixelcoordinates(a,b):#Calculate pixel coordinates (a,b) (3)# (4) If(I(a,b)==0&B(x,y)==1):#If (a,b) in the I image is equal to 0 and (x,y) in the B image is equal to 1, then f is True setftoTrue. End If(f==True): settheoutputpixelO(i,j)to1.#Set the output pixel O(i,j) to 1 End Else: settheoutputpixelO(i,j)to0.#Set the output pixel O(i,j) to 0 End End End Remove the padding from the output image O#Remove the padding from the output image O and return the output image O.

[0080] In one example, the above-mentioned original image boundary alignment preprocessing method may include: equal height alignment processing, which can form a maximum contour by extracting the most obvious points of the original image boundary, which is conducive to more accurate image segmentation.

[0081] Furthermore, the above-mentioned algorithm for processing the same height may include the following contents: Input: image I of size mxn Output: image O of size mxn Foreachpixel(i,j)intheinputimageI:#For each pixel (i,j) in the input image I Computethebiggestcontour#Calculate the largest contour line

[0082] Computetheextremepoints#Calculate extreme values leftmost=(image,8,(0,0,255),-1) rightmost=(image,8,(0,255,0),-1) topmost=(image,8,(255,0,0),-1) bottommost=(image,8,(255,255,0),-1) End Draw the contour and extreme points in the image I # Draw the extreme points and contours in the image I Croptheimage#Returns the cropped image.

[0083] Return the cropped image O It is understandable that not all of the above-mentioned original image preprocessing methods need to be executed. In the case where all of the above-mentioned methods are executed, the execution order is: original image denoising preprocessing - original image data preprocessing - original image homogenization preprocessing - original image target data expansion preprocessing - original image boundary alignment preprocessing.

[0084] Example 4: Combining the above example 1 and Figure 3 Assuming that the above target biological image is an image representing human chest tissue, the image can be first denoised and smoothed by using the Gaussian blur algorithm (that is, the above original image denoising preprocessing); then the image is converted to grayscale and sharpened by threshold processing (that is, original image data preprocessing), thereby simplifying the image for further processing and segmenting the difficult-to-segment features in the image; then, the image is eroded (that is, original image homogenization preprocessing) to facilitate the classification of similar features in the image to the greatest extent and reduce the amount of subsequent processing data; then, the image is amplified (original image target data expansion preprocessing) to facilitate further amplification of abnormal factor features in the image; finally, the image is aligned with equal heights (that is, original image boundary alignment preprocessing), so that the original image can be flipped vertically and horizontally to help the subsequent model better represent the changes in the chest position represented in the original image.

[0085] In another embodiment of the present application, a specific implementation method for pre-segmenting the target biological image and generating the target biological features is also provided. Exemplarily, the above-mentioned "the above-mentioned biological object calibration segmentation processing of the target biological image through the segmentation model to obtain the target biological features" includes: performing resolution compression processing on the above-mentioned target biological image through the segmentation model to generate a first biological image of a first resolution; extracting the initial contour information of the above-mentioned biological object in the above-mentioned first biological image through the segmentation model; comparing the image information at the position corresponding to the above-mentioned initial contour information in the above-mentioned first biological image through the segmentation model to generate updated contour information, and confirming the above-mentioned target biological features according to the above-mentioned updated contour information.

[0086] Exemplarily, the first resolution is information that is smaller than a preset resolution threshold.

[0087] Exemplarily, the above-mentioned preset resolution is system-customized or preset, and the embodiments of the present application are not limited to this.

[0088] Exemplarily, the updated profile information is the refined information of the initial profile information.

[0089] It can be understood that in order to improve the problems of excessive image data, overly complex model processing, long time and easy errors that existed before the embodiments of the present application, in the embodiments of the present application, the resolution of the target biological image is first compressed to reduce the amount of data processing, and the target biological image with reduced data volume is processed by the segmentation model.

[0090] Furthermore, in the process of identifying the target biological features in the target biological image, the target biological features have boundary features. The segmentation model can be trained in advance with a large amount of low-resolution data to identify the target biological features in advance, thereby ensuring that in this step, after the segmentation model identifies the target biological image, the boundary recognition of the target biological features at the first resolution is correct, and the recognition can be completed as quickly as possible.

[0091] Exemplarily, the initial contour information may be a boundary of the target biological feature at the first resolution.

[0092] Exemplarily, after the initial contour information is determined, a refinement process can be performed on the initial contour information, that is, by comparing the image data of the target biological image, a high-resolution image, with the image data at the position corresponding to the initial contour information, thereby refining and updating the initial contour information.

[0093] It is understandable that in the subsequent process of comparing the target biological image with the image data at the corresponding position of the initial contour information, the image range that the segmentation model needs to compare in the target biological image has been greatly reduced, that is, the amount of data comparison has been reduced. Therefore, this solution can significantly improve the efficiency of target biological feature segmentation while ensuring the correct segmentation of target biological features.

[0094] Optionally, in an embodiment of the present application, the above-mentioned generation of updated contour information by comparing the image information at a position corresponding to the above-mentioned initial contour information in the above-mentioned target biological image through the segmentation model includes: controlling the resolution of the above-mentioned first biological image to increase at a preset resolution change rate through the above-mentioned segmentation model to generate multiple second sub-biological images; according to the above-mentioned multiple second sub-biological images, updating the image information at a position corresponding to the above-mentioned initial contour information in the above-mentioned first biological image, and generating updated contour information.

[0095] Exemplarily, the preset resolution change rate is: a growth rate from the first resolution to the resolution of the target biological image within a preset time period, and a resolution of one second biological sub-image corresponding to a moment.

[0096] It can be understood that in the embodiment of the present application, the process of refining the above-mentioned contour information by the above-mentioned segmentation model is completed gradually, that is, it continuously controls the resolution of the above-mentioned first biological image to increase at a preset rate, so that the above-mentioned first biological image is continuously updated, and the image information at the position corresponding to the initial contour information is continuously refined, that is, multiple second sub-biological images will be generated, and the resolution of each second sub-biological image is higher than the resolution of the first biological image. After that, the segmentation model will continuously compare the updated second sub-biological image with the second sub-biological image or the first biological image before this update, and then continuously refine it until it is updated to the target biological image, the contour information refinement is completed, and the final finer contour information is generated.

[0097] It can be understood that the above segmentation model can be a U-net network model, and the U-net network model is introduced as follows: U-Net is a semantic segmentation network developed on the basis of fully convolutional neural networks. The network has 23 layers and belongs to the feature pyramid network, which has significantly fewer layers than other networks while maintaining accuracy. The feature pyramid network is used to extract feature vectors from the segmented image. This feature vector is used as input to the neural network classifier for the unlabeled nodule prediction task. Contour analysis isolates objects in the image by drawing lines around the image. The segmentation of the U-Net network has two sides: one is contracting and the other is symmetrically expanding. Convolutional layers, linear units (ReLU), dropout layers, and pooling layers partially constitute the contracting side. When downsampling, the number of feature channels increases by 2 times. Parts of the upsampling (transposed convolution) layers, connections to the relevant feature channels of the contracting side, convolutional layers, and dropout layers are included in the expanding side. The network is able to use the connections between the contracting and expanding sides to obtain the required features from the relevant layers. In each region of the expanding side, the number of feature channels is halved. The contracting path at the end is intended to capture the contextual information and high-level features of the image. On the other hand, the decoder network is the expanding part of the architecture. It consists of upsampling layers and convolutional layers, which gradually improve the spatial resolution of the biological images whose features have been identified while reducing the number of feature channels.

[0098] This expansion path aims to reconstruct high-resolution details of the image and refine the segmentation mask. The encoder and decoder networks are connected by a bottleneck layer that preserves the high-resolution features of the identified biological images. The jump between the encoder and decoder networks helps to combine the high-resolution features from the encoder with the low-resolution features from the decoder, resulting in more accurate segmentation.

[0099] When the network model segments biological images that are already at low resolution and have blurred boundaries, U-Net combines low-resolution and high-resolution information through skip connections. Low-resolution information is used for target recognition, and high-resolution information is used for positioning and segmentation. The contraction path efficiently extracts and compresses features through alternating convolution and pooling operations, allowing the subsequent expansion path to use these high-level features for accurate segmentation. The expansion path mainly uses the gradual restoration of the spatial resolution of the feature map compressed in the contraction path, while combining the high-resolution features of the skip connection to achieve biological image segmentation.

[0100] In another embodiment of the present application, a specific implementation method for pre-segmenting the target biological image and generating the target biological features is also provided. Exemplarily, the above-mentioned "detecting abnormal factors of the above-mentioned target biological features through the detection model to obtain the abnormal detection results of the above-mentioned biological objects to be identified" includes: determining multiple feature information corresponding to each abnormal factor in the above-mentioned abnormal factors through the detection model; according to the multiple feature information of each abnormal factor, identifying the regional features in the above-mentioned target biological features that match each abnormal factor as the abnormal detection results of the above-mentioned biological objects to be identified.

[0101] Exemplarily, the above-mentioned characteristic information is used to indicate different manifestations of each abnormal factor.

[0102] For example, it can be known from the above content that each of the above abnormal factors, such as abnormal size of biological objects, abnormal morphology of biological objects, and abnormal absorption characteristics of biological objects to radiation energy, has its own different characteristic information. For example, for abnormal size of biological objects, it may have multiple sizes that all belong to abnormal size of biological objects. These different characteristic information are obtained after training. When using the detection model to detect these abnormal factors, it is necessary to identify and compare the multiple different characteristic information of each abnormal factor, and the matching image areas all belong to the abnormal detection results.

[0103] Furthermore, assuming that the abnormal detection result is a lung nodule feature, the lung nodule feature may have a size of 1 cm, 2 cm, 3 cm, 4 cm, or 5 cm, which are all abnormal sizes of biological objects; at the same time, in terms of morphological features, it may have a hexagonal, octagonal, surface burrs, and other shapes, which are all abnormal shapes of biological objects. Based on this, it is necessary to arrange and combine these abnormal factors, and all combinations need to be identified in order to comprehensively screen and identify the abnormal detection results.

[0104] Optionally, in an embodiment of the present application, when the abnormal factors are abnormal sizes of biological objects, and include multiple levels of abnormal sizes of biological objects, the above-mentioned detection model is used to determine the multiple feature information corresponding to each of the abnormal factors, and the aforementioned “determine the multiple feature information corresponding to each of the abnormal factors through the detection model” includes: determining the multiple size information corresponding to the abnormal sizes of the biological objects through the detection model. The aforementioned “identifying the regional features in the target biological features that match each of the abnormal factors as the abnormal detection results of the biological objects to be identified based on the multiple feature information of each of the abnormal factors” includes: identifying the regional features in the target biological features that match the multiple size information as the abnormal detection results of the biological objects to be identified.

[0105] As mentioned above, the detection model has already included the feature information corresponding to each abnormal factor in the previous training. Therefore, the detection model is first used to determine the multiple size information corresponding to the abnormal size of the above-mentioned biological object. Then, based on these size information, it is identified in the image corresponding to the target biological feature. The area in the image corresponding to these size information is the abnormal detection result.

[0106] It can be understood that from the foregoing content, the detection model can be based on a model using a ResNet network as the backbone and a feature pyramid (FPN) network as the architecture. The following describes this model: The FPN network is a feature extraction network that enhances the performance of CNN by utilizing multi-scale feature maps, so that the network can recognize objects of different scales. At the same time, in the embodiment of the present application, the ResNet network is a deep residual network architecture, which can run deeper network depth without encountering the gradient vanishing problem.

[0107] The FPN network is designed to detect problems of different scales in biological images and extract structures from input biological images. This refers to extracting features by combining upsampling of lower-resolution feature maps with higher-resolution feature maps. Simply put, feature extraction is performed on images of each scale, which can produce multi-scale feature representations, and all levels of feature maps have strong semantic information, including high-resolution feature maps.

[0108] Optionally, in an embodiment of the present application, the recognition method based on the biological image recognition model provided by the present application also includes a specific implementation method of training and generating the above-mentioned biological image recognition model. Specifically, the above-mentioned model includes a segmentation model, a detection model, and a feature determination model. Accordingly, for different models, corresponding training processes need to be carried out.

[0109] Exemplarily, with respect to the segmentation model, the method further includes: acquiring a plurality of original biological images and biological features corresponding to the plurality of original images; inputting the plurality of original biological images into the segmentation model to generate a plurality of first test arrays corresponding to a plurality of groups of original biological images; comparing the plurality of first test data with the biological features corresponding to the plurality of original images, and determining a first difference between the plurality of first test data and the biological features corresponding to the plurality of original images; and when the difference is greater than a first preset threshold, adjusting the segmentation model according to the first difference, so that the first difference between the first test data generated by the segmentation model and the biological features corresponding to the original images is less than a first preset threshold.

[0110] Exemplarily, there is a one-to-one correspondence between the multiple original biological images and the biological features corresponding to the multiple original images.

[0111] Exemplarily, the first preset threshold may be preset in advance or may be user-defined.

[0112] Exemplarily, the first preset threshold can be used to indicate whether the training results of the segmentation model meet user requirements. When the difference between the first test data and the biometric features corresponding to the original image is less than the first preset threshold, the training can be stopped to generate a qualified segmentation model. Otherwise, it is necessary to input multiple sets of image relationships again for training until the first difference between the first test data and the biometric features is less than the first preset threshold.

[0113] It is understandable that after the segmentation model is constructed, it is necessary to train the segmentation model so that the segmentation model can accurately complete the process of identifying the target biological features in subsequent target biological images.

[0114] Furthermore, multiple sets of original images for training and image relationships of biometric features corresponding to the original images may be preset in advance, and then the multiple sets of image relationships may be input into the segmentation model to generate multiple sets of first test arrays.

[0115] It is understandable that during the training process, the first difference between the generated first test array and the biometric features corresponding to the original image that meets user requirements may be greater than the first preset threshold. In this case, it is necessary to input multiple sets of image relationships into the segmentation model again and continue training until the difference between the first test data and the biometric data is less than the first preset threshold.

[0116] Exemplarily, with respect to the detection model, the method further includes: obtaining multiple biometric features and abnormality detection results corresponding to the multiple biometric features; inputting the multiple biometric features into the detection model to generate multiple first test arrays corresponding to the multiple biometric features; comparing the multiple first test data with the abnormality detection results corresponding to the multiple biometric features, and determining a second difference between the multiple first test data and the abnormality detection results corresponding to the multiple biometric features; when the second difference is greater than a second preset threshold, adjusting the detection model according to the second difference so that the second difference between the second test data generated by the detection model and the abnormality detection result corresponding to the original image is less than a second preset threshold.

[0117] Exemplarily, there is a one-to-one corresponding second image relationship between the multiple biometric features and the abnormality detection results corresponding to the multiple biometric features.

[0118] Exemplarily, the second preset threshold may be preset in advance or may be user-defined.

[0119] Exemplarily, the second preset threshold can be used to indicate whether the training results of the detection model meet user requirements. When the difference between the second test data and the abnormal detection result of the biometric feature is less than the second preset threshold, the training can be stopped to generate a qualified detection model. Otherwise, it is necessary to input multiple sets of image relationships again for training until the second difference between the second test data and the target data is less than the second preset threshold.

[0120] It is understandable that after the detection model is constructed, it is necessary to train the detection model so that the detection model can accurately complete the process of identifying abnormal detection results in subsequent target biometric features.

[0121] Furthermore, multiple sets of image relationships between biometric features for training and abnormality detection results corresponding to the biometric features may be preset in advance, and then multiple sets of second image relationships may be input into the segmentation model to generate multiple sets of second test arrays.

[0122] It is understandable that during the training process, the second difference between the generated second test array and the biometric features corresponding to the original image that meets user needs may be greater than the second preset threshold. In this case, it is necessary to input multiple sets of second image relationships into the detection model again and continue training until the second difference between the second test data and the anomaly detection result is less than the second preset threshold.

[0123] Exemplarily, with respect to the feature determination model, the method further includes: obtaining abnormality detection results corresponding to multiple biometric features and properties of abnormal factors corresponding to the multiple abnormality detection results; inputting the abnormality detection results corresponding to the multiple biometric features into the feature determination model to generate multiple third test arrays corresponding to the abnormality detection results corresponding to the multiple biometric features; comparing the multiple third test data with the properties of the abnormal factors corresponding to the multiple abnormality detection results, and determining a third difference between the multiple second test data and the properties of the multiple abnormal factors; when the third difference is greater than a third preset threshold, adjusting the feature determination model according to the third difference, so that the third difference between the third test data generated by the feature determination model and the properties of the abnormal factors is less than a third preset threshold.

[0124] Exemplarily, there is a one-to-one correspondence between the multiple abnormality detection results and the properties of the abnormal factors corresponding to the multiple abnormality detection results.

[0125] Exemplarily, the third preset threshold value may be preset in advance or may be user-defined.

[0126] Exemplarily, the third preset threshold can be used to indicate whether the training results of the feature judgment model meet user requirements. When the third difference between the properties of the abnormal factors corresponding to the third test data and the abnormal detection results is less than the third preset threshold, the training can be stopped to generate a qualified detection model. Otherwise, it is necessary to input multiple sets of third image relationships again for training until the third difference between the properties of the abnormal factors corresponding to the third test data and the abnormal detection results is less than the third preset threshold.

[0127] It is understandable that after the feature determination model is constructed, it is necessary to train the feature determination model so that the feature determination model can accurately complete the process of identifying the nature of the abnormal factors corresponding to the subsequent abnormal detection results.

[0128] Furthermore, a plurality of third test arrays can be generated by presetting in advance a third image relationship between the abnormal detection results corresponding to a plurality of biometric features and the nature of the abnormal factors corresponding to the plurality of abnormal detection results, and then inputting the plurality of sets of third image relationships into a feature determination model.

[0129] It is understandable that, during the training process, the third difference between the generated third test array and the biometric features corresponding to the original image that meets user requirements may be greater than the third preset threshold. In this case, it is necessary to input multiple sets of third image relationships into the feature determination model again and continue training until the third difference between the third test data and the nature of the abnormal factor corresponding to the abnormal detection result is less than the third preset threshold.

[0130] In an embodiment of the present application, in the process of training the above-mentioned feature determination model, the LIDC-IDRI database can be used to train the model, and the prediction model is more applicable.

[0131] We will test the proposed algorithm on the LIDC dataset to see if it can accurately classify and identify specific features, and compare the results with standard metrics such as precision, recall, and F1 score. The Adam optimizer is added to stabilize the model performance and reduce the impact of errors.

[0132] The present application also provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the training rule determination method described in the present application is implemented. For example, the following can be executed: Figure 1 The individual steps of the method are shown.

[0133] The present invention provides a computer program product, which includes instructions that are executed by a processor when the processor executes the instructions. Figure 1 The individual steps of the method are shown.

[0134] It should be noted that although the operations of the method of the present invention are described in a particular order in the drawings, this does not require or imply that the operations must be performed in this particular order or that all illustrated operations must be performed to achieve desired results.

[0135] Figure 4 This is a block diagram of a recognition device based on a biological image recognition model according to an embodiment of the present application. Figure 4 , the device includes an execution module 601.

[0136] An execution module, used to perform biological object calibration and segmentation processing on the target biological image through a segmentation model to obtain a target biological feature, wherein the target biological feature is used to characterize the biological object to be identified in the target biological image; The execution module is further used to perform abnormal factor detection on the target biological feature through the detection model to obtain an abnormal detection result of the biological object to be identified; the abnormal detection result includes whether the biological object to be identified has an abnormal factor; the abnormal factor includes at least one of an abnormal size of the biological object, an abnormal morphology of the biological object, and an abnormal absorption characteristic of the biological object to radiation energy; The execution module is also used to determine the properties of the abnormal factors of the target biological features through a feature determination model to obtain the properties of the abnormal factors, and the properties are used to characterize whether there is a risk in the abnormal factors and the risk level of the abnormal factors.

[0137] In one embodiment, the execution unit 601 is also used to obtain original image data, perform image preprocessing on the original image data, and generate a target biological image; the image preprocessing includes at least one of the following: original image denoising preprocessing, original image data preprocessing, original image homogenization preprocessing, original image target data expansion preprocessing, and original image boundary alignment preprocessing.

[0138] In one embodiment, the execution unit 601 is specifically used to perform resolution compression processing on the target biological image through a segmentation model to generate a first biological image with a first resolution, where the first resolution is information less than a preset resolution threshold; extract initial contour information of the biological object in the first biological image through the segmentation model; compare the image information at a position corresponding to the initial contour information in the target biological image through the segmentation model to generate updated contour information, and confirm the target biological feature based on the updated contour information, where the updated contour information is refined information of the initial contour information.

[0139] In one embodiment, the execution unit 601 is specifically used to control the resolution increase of the first biological image according to a preset resolution change rate through the segmentation model to generate multiple second sub-biological images, wherein the preset resolution change rate is: the growth rate of the resolution from the first resolution to the target biological image within a preset time period, and the second sub-biological image corresponds to the resolution at a moment; according to the multiple second sub-biological images, the image information of the position corresponding to the initial contour information in the first biological image is updated, and the updated contour information is generated.

[0140] In one embodiment, the execution unit 601 is specifically used to determine, through a detection model, a plurality of feature information corresponding to each of the abnormal factors, wherein the feature information is used to indicate different manifestations of each abnormal factor; and based on the plurality of feature information of each abnormal factor, identify the regional features in the target biological features that match each abnormal factor as the abnormality detection result of the biological object to be identified.

[0141] In one embodiment, the execution unit 601 is specifically used to determine multiple size information corresponding to the abnormal size of the biological object through a detection model; and identify regional features in the target biological features that match the multiple size information as abnormal detection results of the biological object to be identified.

[0142] It should be understood that the units recorded in the recognition device based on the biological image recognition model correspond to the various steps in the method described in the accompanying drawings. Therefore, the operations and features described above for the method are also applicable to the recognition device based on the biological image recognition model and the units contained therein, and will not be repeated here. The recognition device based on the biological image recognition model can be pre-implemented in the browser or other security applications of the computer device, and can also be loaded into the browser or its security application of the computer device by downloading or the like. The corresponding units in the recognition device based on the biological image recognition model can cooperate with the units in the computer device to implement the solution of the embodiment of the present application.

[0143] For the several modules or units mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0144] It should be noted that for details not disclosed in the recognition device based on the biological image recognition model in the embodiment of the present application, please refer to the details disclosed in the above embodiments of the present application, and no further details will be given here.

[0145] Reference below Figure 5 , Figure 5 FIG. 1 is a schematic diagram showing a structure of a computer device suitable for implementing an embodiment of the present application. Figure 5 As shown, the computer system 1700 includes a central processing unit (CPU) 1701, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1702 or the program loaded from the storage part 1708 to the random access memory (RAM) 1703. In the RAM 1703, various programs and data required for the operation instructions of the system are also stored. The CPU 1701, the ROM 1702, and the RAM 1703 are connected to each other through a bus 1704. An input / output (I / O) interface 1705 is also connected to the bus 1704.

[0146] The following components are connected to the I / O interface 1705: an input section 1706 including a keyboard, a mouse, etc.; an output section 1707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1708 including a hard disk, etc.; and a communication section 1709 including a network interface card such as a LAN card, a modem, etc. The communication section 1709 performs communication processing via a network such as the Internet. A drive 1710 is also connected to the I / O interface 1705 as needed. A removable medium 1711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1710 as needed, so that a computer program read therefrom is installed into the storage section 1708 as needed.

[0147] In particular, according to an embodiment of the present application, the above reference flow chart Figure 1 The described process can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication part 1709, and / or installed from a removable medium 1711. When the computer program is executed by the central processing unit (CPU) 1701, the above-mentioned functions defined in the system of the present application are executed.

[0148] It should be noted that the computer-readable medium shown in the present application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium such as a computer-readable storage medium that can send, propagate or transmit a program for use by or in conjunction with an instruction execution system, device or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0149] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, functions and operating instructions of the system, method and computer program product according to various embodiments of the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the aforementioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, the boxes represented by two connections can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operating instruction, or can be implemented with a combination of dedicated hardware and computer instructions.

[0150] The units or modules involved in the embodiments described in the present application may be implemented by software or hardware. The units or modules described may also be set in a processor. For example, it may be described as: a processor includes a first receiving module, a second receiving module, and a sending module. The names of these units or modules do not, in some cases, constitute limitations on the units or modules themselves.

[0151] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiment, or may exist independently without being assembled into the electronic device. The above computer-readable storage medium stores one or more programs, and when the above programs are used by one or more processors to execute the recognition method based on the biological image recognition model described in the present application.

[0152] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) by each other to form a technical solution.

Claims

1. A recognition method based on a biological image recognition model, characterized in that: include: Performing biological object calibration and segmentation processing on the target biological image through the segmentation model to obtain target biological features, wherein the target biological features are used to characterize the biological object to be identified in the target biological image; The target biological feature is subjected to abnormal factor detection through the detection model to obtain an abnormal detection result of the biological object to be identified; the abnormal detection result includes whether the biological object to be identified has abnormal factors; the abnormal factors include at least one of abnormal size of the biological object, abnormal morphology of the biological object, and abnormal absorption characteristics of the biological object to radiation energy; The abnormal factors of the target biological features are judged by the feature judgment model to obtain the properties of the abnormal factors, which are used to characterize whether there is a risk in the abnormal factors and the risk level of the abnormal factors.

2. The method according to claim 1, characterized in that: Before performing biological object calibration and segmentation processing on the target biological image through the segmentation model to obtain the target biological features, the method includes: Acquiring original image data, performing image preprocessing on the original image data, and generating a target biological image; The image preprocessing includes at least one of the following: original image denoising preprocessing, original image data preprocessing, original image homogenization preprocessing, original image target data expansion preprocessing, and original image boundary alignment preprocessing.

3. The method according to claim 1, characterized in that The step of performing biological object calibration and segmentation processing on the target biological image by using the segmentation model to obtain the target biological features includes: Performing resolution compression processing on the target biological image through a segmentation model to generate a first biological image with a first resolution, wherein the first resolution is information less than a preset resolution threshold; extracting initial contour information of the biological object in the first biological image by using a segmentation model; The image information at the position corresponding to the initial contour information in the target biological image is compared with the segmentation model to generate updated contour information, and the target biological feature is confirmed according to the updated contour information, wherein the updated contour information is the refined information of the initial contour information.

4. The method according to claim 3, characterized in that The step of comparing the image information at a position corresponding to the initial contour information in the target biological image with the segmentation model to generate updated contour information comprises: Controlling the resolution of the first biological image to increase according to a preset resolution change rate through the segmentation model to generate a plurality of second sub-biological images, wherein the preset resolution change rate is: the growth rate from the first resolution to the resolution of the target biological image within a preset time period, and one second sub-biological image corresponds to a resolution at one moment; According to the plurality of second sub-biological images, image information of a position corresponding to the initial contour information in the first biological image is updated, and updated contour information is generated.

5. The method according to claim 1, characterized in that The detecting of abnormal factors on the target biological feature by using a detection model to obtain an abnormal detection result of the biological object to be identified includes: Determine, by means of a detection model, a plurality of feature information corresponding to each of the abnormal factors, wherein the feature information is used to indicate different manifestations of each abnormal factor; According to the plurality of feature information of each abnormal factor, the regional features in the target biological features that match each abnormal factor are identified as the abnormality detection result of the biological object to be identified.

6. The method according to claim 5, characterized in that In the case where the abnormal factor is an abnormal size of a biological object and includes multiple levels of abnormal sizes of biological objects, the detection model is used to determine multiple feature information corresponding to each abnormal factor in the abnormal factor, including: Determining, by means of a detection model, a plurality of size information corresponding to the abnormal size of the biological object; The step of identifying, based on the plurality of feature information of each abnormal factor, the regional features in the target biological features that match each abnormal factor as the abnormality detection result of the biological object to be identified includes: The regional features in the target biological features that match the multiple size information are identified as the abnormality detection results of the biological object to be identified.

7. A recognition device based on a biological image recognition model, characterized in that: include: An execution module, configured to perform biological object calibration and segmentation processing on a target biological image through a segmentation model to obtain a target biological feature, wherein the target biological feature is used to characterize a biological object to be identified in the target biological image; The execution module is further used to perform abnormal factor detection on the target biological feature through the detection model to obtain an abnormal detection result of the biological object to be identified; the abnormal detection result includes whether the biological object to be identified has an abnormal factor; the abnormal factor includes at least one of an abnormal size of the biological object, an abnormal morphology of the biological object, and an abnormal absorption characteristic of the biological object to radiation energy; The execution module is also used to determine the properties of the abnormal factors of the target biological features through a feature determination model to obtain the properties of the abnormal factors, and the properties are used to characterize whether the abnormal factors have risks and the risk level of the abnormal factors.

8. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising instructions, characterized in that: When the instructions are executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images